CGNICO/UnrealEngine
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
c11d382a6f0acddb0bc1e95ab9bc5f8fc3570b55
UnrealEngine/Engine/Source/ThirdParty/OpenVDB/openvdb-12.0.0/tsc/meetings/2019-06-20.md
Jeffreytsai1004 c11d382a6f ue5-main
2025-05-18 13:04:45 +08:00

4.3 KiB
Raw Blame History

Minutes from 21st OpenVDB TSC meeting, June 20, 2019

Attendees: Nick A., Dan B., Peter C., Jeff L., Ken M.

Additional Attendees: Bruce Chernia (Intel)

Agenda:

  1. Quorum

  2. Secretary

  3. Siggraph planning

  4. PagedArray unit test fix

  5. SOP unification and hiding

  6. SOP tab sub-menu and suffix

  7. Static Pointer Memory Leaks (PR 227) 7a) CI Update 7b) Houdini status

  8. Memory Allocators

  9. VDB Points MMB bug (OVDB-100)

  10. Speed of hasActiveTiles/voxelizeActiveTiles

  11. GCC and Dual ABI

  12. Next Meeting

  13. Quorum was confirmed.

  14. Secretary was Jeff Lait

  15. Siggraph planning

We have a dedicated meeting slot in the BoF for discussing development issues rather than as an open forum. The plan is Wednesday 11am-Noon to discuss with Autodesk the MultiResolution issues.

With both a course and a BoF, it is unclear how to handle the BoF. Less need to just dump new updates as the course will cover that. "What would you like to see?" was a goo conversation starter in the past. Previous answers were Multi Res and Delayed Loading.

  1. PagedArray unit test fix

Still trying to build a PR for this. Unfortunately, git is very painful to work with. Might end up just being a patch forwarded. gitk is listed as a good tool to figure out what git is trying to do.

  1. SOP unification and hiding

We now have separate PRs for both adding spare data to OP_Operator and for creating a pythonrc to do the hiding. Consensus is both the env-variable and pythonrc can go in as the overlap has been minimized. It would be a good if it triggered a warning if you turned both on.

The SpareData technique should be adopted by Houdini some day to allow generic sparedata on all OP_Operators. RFE: 97588 has been submitted for this.

The environment variable should match the pythonrc, so be a three way toggle.

CMake should let you set both, but with a warning that you likely will confuse yourself.

  1. SOP tab sub-menu and suffix

Previous default menu was Custom. Instead, by using a named menu we can ensure conflicts get a suffix automatically. ASWF isn't a good top level menu as no one will know to look in it. Renaming nodes is finicky, so consensus is the tab suffixing is a better approach to deal with ambiguity.

Bug: 97465 for suffixes not working has been fixed in 17.0.620, 17.5.295.

  1. Static Pointer Memory Leaks (PR 227)

This is a long standing PR from Autodesk. We should at least answer.

It wasn't merged because older versions of MSVC did not conform to C++11 for function statics. It also isn't a memory leak, as it will be freed on destruction in either case. Further, destruction order causes its own risks so safest to not delete. OTOH, C++11 is now a requirement and going forward we can support function statics.

Next version will be C++14 so this implies 2017 MSVC. We should state this on the upgrade.

While this PR isn't necessary, it would be a good test for the new CLA process. Jeff will comment on the PR and ask them to wait for the CLA mechanism to be active.

7a) CI Update

Switching over to a host repository on a ASWF docker. This will have a pre-installed CentoS7 with all the dependencies pre-built. All is working except the Houdini install. Installing glfw is problematic, however. Bumping glfw (which is only used by vdbview) to a higher version will likely simplify. Unanimous consent to bump to 3.1 to match Ubunut 16.04.

Minimum is currently 3.0. But 3.1 is what you get with apt-get. However, vdbview crashes on close with unpatched glfw due to EGL problems. Recommendation is 3.3.

7b) Houdini status

Attempt to upgrade HEAD Houdini to latest VDB ran into a problem with the VDB Clip changes. The question of if primitive attributes or vdb metadata should be authoritative when processing SOPs (and when processing VDB code) is thorny and error prone. In this case, file_bbox is generated on save to .bgeo streams. So reloaded .vdbs have the value as metadata and do not have matching primitive and vdb data. Interestingly, file_bbox is off in the open source build. Should it be on by default in all builds?

  1. Memory Allocators

Discussion at Siggraph on memory allocators and file format changes.

  1. VDB Points MMB bug (OVDB-100)

Jeff Lait is to look at this.

  1. Speed of hasActiveTiles/voxelizeActiveTiles Time

  2. GCC and Dual ABI Time

  3. Schedule Next Meeting

June 27th 2019. 3pm-4pm EDT (GMT-4).